Agent tool protocols and checkout standards are arriving fast. The hard part is having true inventory behind them. Here's where the protocols fit, and where JENNi fits.
If you follow agentic commerce protocols, you've seen the pattern. Every few weeks there's a new proposal for how an agent should discover products, start an order or pay. Most of them are sensible. None of them can answer a simple question: is the thing actually there?
What agentic commerce protocols do
Two kinds of interface matter for agents that shop.
Tool protocols let an agent discover and call capabilities. The one we use is Model Context Protocol, or MCP. A server describes its tools, such as search, product details and buying options, and any agent that speaks MCP can use them without custom integration code.
Checkout-session standards describe the order itself: how a session is opened, how items, delivery and totals are updated, and how payment is completed. They give agents and sellers a shared shape for the steps between "I want this" and "it's ordered."
Both are plumbing. Good plumbing, and worth having. But plumbing carries what you put in it.
A protocol doesn't create inventory or truth
A protocol can say how to ask for a product. It can't make the product exist.
If the data behind the interface is a stale feed, the agent gets a clean, well-formatted wrong answer. Stock says yes and the shelf is empty. A delivery promise is polite and impossible. The standard worked exactly as designed, and the shopper still lost.
This is why the data layer decides whether agentic commerce works. Someone has to know what's on physical shelves, how sure they are right now, and what a courier can really do for a given address. Without that, protocols just move guesses faster.
How JENNi exposes shelf-backed products
JENNi makes inventory on physical retail shelves usable by digital demand: shoppers, apps and AI agents. The SKU Graph is the shared layer underneath. It joins canonical products anchored on barcodes with the stores that carry them, including price, stock, source and when we last saw each offer, plus how sure we are right now.
Our MCP server at https://shop.jennipro.com/mcp puts that behind plain tools. search_catalog finds in-stock items near a ZIP. get_buying_options returns availability, price, delivery fee, delivery promise and total before tax. start_checkout hands off to a secure hosted page, and create_checkout, update_checkout and complete_checkout run checkout in chat with a one-time payment token.
Same day is promised only when the quote says "Get it today." Checkout refuses stale prices. Delivered orders and failed orders both feed back into the graph, so what the next agent sees is shaped by what actually happened.
The interface is the easy part. The graph behind it is the work.
Why the data layer is the slow part
Writing a standard takes months. Building a trustworthy picture of what's on shelves takes longer, because it never stops. Prices change, stock moves, and every order that succeeds or fails teaches the graph something. That feedback is the part a specification can't supply, and it's why we treat the data as the product and the interface as a door into it.
What to watch as standards settle
A few things are worth watching.
- Convergence. Several standards may merge, or one may win. Build your agent so the commerce layer is a tool it calls, not logic baked into its prompt. Then you can adopt another interface without rewriting your agent.
- Where trust lives. Expect pressure to answer "who is the seller?" and "who handles problems?" clearly. On JENNi orders, JENNi is the seller, so issues come to us.
- Payment handling. The safest designs keep card numbers out of agent conversations entirely. Hosted pages and one-time tokens are the pattern to prefer.
- Truthful promises. The standards that matter will be the ones that make it easy to say "not available" and "can't be same day" clearly.
Where we come in
Standards will keep changing. Shelves will still be where the goods are. If you're building an agent, connect to https://shop.jennipro.com/mcp and see what an interface looks like with real inventory behind it. And if you run retail shelves: want your shelves visible to AI agents? Get in touch at jennipro.com.